McGraw-Hill Education Copyright 2017 by McGraw-Hill Education. All rights reserved.

FORECASTING
Chapter 3
Forecasting in Operations and Supply
Chain Management
q Forecasting is a vital function and impacts every
significant management decision
¤ Finance and accounting use forecasts as the basis for
budgeting and cost control
¤ Marketing relies on forecasts to make key decisions
such as new product planning and personnel
compensation
¤ Production uses forecasts to select suppliers, determine
capacity requirements, and to drive decisions about
purchasing, staffing, and inventory

3- 2
Decoupling Points
q Decoupling points occur when inventory is positioned
in the supply chain to allow one operation to act
independently of another
¤ For example, inventory at a retail store separates
(buffers) the manufacturer from the actions of individual
consumers
q Forecasts of demand at these decoupling points
allows inventory to be set to the proper level

3- 3
 LO2

Demand Management

Independent Demand:
Finished Goods

A Dependent Demand:
Raw Materials,
Component parts,
B(4) C(2) Sub-assemblies, etc.

D(2) E(1) D(3) F(2)

3- 4
 Forecasting Methods
5

Forecasting
Methods

Quantitative Qualitative

Causal Time Series

Types of Forecasting
q Four basic types of forecasts are common
1. Quantitative
i. Time series analysis
ii. Causal relationships
iii. Simulation
2. Qualitative
q Time series analysis is based on the idea that data
relating to past demand can be used to predict
future demand

3- 6
7

Quantitative forecasting methods

Time series models Causal models

¨ Past predicts future ¨ Examines potential cause --> effect

¨ Uses time series data

relationships
¨ Uses cross sectional or time series
¨ Key variable: time (t)
data
¨ Easier to apply
¨ Key variables are usually denoted as
¨ Less accurate X1 , X 2 , X 3 , …
¨ Examples: ¨ More difficult and takes more time

¤ Moving averages ¨ But worth it, because it provides

insight to the system (process) under
¤ Exponential smoothing
study
¨ Example: regression

3- 7
 What to look for in a time series?
8

q Trend - long-term movement in data ( growth in a business)

q Seasonality - short-term regular and repetitive variations in
data
q Cyclical variations – long(er) term, occasionally caused by
unusual circumstances, (war, economic downturn, etc.)
q Autocorrelation – denotes persistence of occurrence
(momentum driven)
q Random variations - caused by chance

q See Components of Demand.xls

3- 8
 Finding Components of Demand

Seasonal variation

x Linear
x x
x x
x x x Trend
Sales

x
x x x
x
x
xx
x xx x x
x
x
x x x x x x
x x x x x x
x x x
x xxxxx
x
x x

1 2 3 4
Year
3- 9
Trends
q Identification
of trend lines is a common starting
point when developing a forecast
q Common trend types include linear, S-curve,
asymptotic, and exponential : we focus on linear

3- 10
Classification of Time Series Models

q No Trend; No Seasonality
¤ Moving Average Techniques
n Simple MA
n Weighted Moving Average
¤ Exponential Smoothing Technique
q With Trend but Not Seasonality
¤ Exponential Smoothing With Trend
¤ Simple Linear Regression (least Squares method)

q With Trend and Seasonality

¤ Additive vs Multiplicative Demand Models

3- 11
Selecting a Forecasting Method

q The underlying pattern in the time series is an important factor

in selecting a forecasting method
q Thus, a time series plot should be one of the first things
developed when trying to determine what forecasting method
to use
q If we see a horizontal pattern, then we need to select a method
appropriate for this type of pattern
q If we observe a trend in the data, then we need to use a
method that has the capability to handle trend effectively
Moving Averages and Exponential Smoothing

q Now we discuss three forecasting methods that are appropriate

for a time series with a horizontal (stationary) pattern:
q Moving Averages

q Weighted Moving Averages

q Exponential Smoothing

q They are called smoothing methods because their objective is

to smooth out the random fluctuations in the time series
q They are most appropriate for short-range forecasts

3- 13
Smoothing Methods- Moving Averages

q The moving average method consists of computing an

average of the most recent n data values for the series
and using this average for forecasting the value of the
time series for the next period

Σ(𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑛 𝑑𝑎𝑡𝑎 𝑣𝑎𝑙𝑢𝑒𝑠) 𝐴!"# + 𝐴!"$ + 𝐴!"% + ⋯ + 𝐴!"&

𝐹! = =
𝑛 𝑛
where Ft= forecast of the time series for period t
q Note that, each observation in the moving average
calculation receives the same weight, 1/n
Also See Simple Models.xls
3- 14
Moving Averages

q To use moving averages to forecast, we must first

select the order n, or number of time series values, to
be included in the moving average
q A smaller value of n will track shifts in a time series
more quickly than a larger value of n
q If more past observations are considered relevant,
then a larger value of n is better
Simple Moving Average Example

A t-1 + A t-2 + A t-3 +...+A t- n

Ft =
Week Demand n
1 650
2 678
Question: What are the 3-week
3 720
4 785
and 6-week moving average
5 859 forecasts for demand?
6 920 Assume you only have 3 weeks
7 850 and 6 weeks of actual
8 758
9 892
demand data for the
10 920 respective forecasts
11 789
12 844
See Examples.xls
3- 16
 17
Calculating the moving averages gives us:

Week Demand 3-Week 6-Week

1 650 F4=(650+678+720)/3
2 678 =682.67
3 720 F7=(650+678+720
+785+859+920)/6
4 785 682.67
=768.67
5 859 727.67
6 920 788.00
7 850 854.67 768.67
8 758 876.33 802.00
9 892 842.67 815.33
10 920 833.33 844.00
11 789 856.67 866.50
12 844 867.00 854.83
3- 17
©The McGraw-Hill Companies, Inc., 2004
 Plotting the moving averages and comparing them shows how
the lines smooth out to reveal the overall upward trend in this
example

1000
900
Demand
Demand

800
3-Week
700
6-Week
600
Note how the 3-
500
Week is smoother
1 2 3 4 5 6 7 8 9 10 11 12 than the Demand,
Week and 6-Week is even
smoother

3- 18
 Weighted Moving Average Formula

While the moving average formula implies an equal

weight being placed on each value that is being averaged,
the weighted moving average permits an unequal
weighting on prior time periods

The formula for the moving average is:

Ft = w 1 A t-1 + w 2 A t- 2 + w 3 A t-3 + ...+w n A t- n

n
wt = weight given to time period “t”
occurrence (weights must add to one)
åw
i=1
i =1

3- 19
Weighted Moving Average Example Data
Question: Given the weekly demand and weights, what is the forecast for the
4th period or Week 4?

Week Demand Weights:

1 650
2 678 t-1 .5
3 720 t-2 .3
4 t-3 .2

Note that the weights place more emphasis on the most recent data,
that is time period “t-1”

3- 20
Weighted Moving Average Example Solution

Week Demand Forecast

1 650
2 678
3 720
4 693.4

F4 = 0.5(720)+0.3(678)+0.2(650)=693.4

3- 21
Weighted Moving Averages
q To use this method we must first select the number of data
values, n, to be included in the average
q Next, we must choose the weight for each of the data values
q The more recent observations are typically given more weight
than older observations
q For convenience, the weights should sum to 1
q Experience and/or trial-and-error are the simplest
approaches

3- 22
Moving Average Methods vs
Exponential Smoothing Methods
q If you are making forecasts for 20 thousand items in a
Wal-Mart and using a 60-day moving average method,
there is a great computational burden? You need to keep
too much data in storage, and need to do too many
calculations.
q Exponential Smoothing technique (which is very easy to
use and understand), resolves this issue (the issue of large
number of calculations and storage requirements) with
surprising accuracy.

3- 23
Exponential Smoothing

q This method is a special case of a weighted moving averages

method; we select only the weight for the most recent
observation
q The weights for the other data values are computed
automatically and become smaller as the observations grow
older
q The exponential smoothing forecast is a weighted average of
all the observations in the time series
q The term exponential smoothing comes from the exponential
nature of the weighting scheme for the historical values
Exponential Smoothing Model

𝐹! = 𝐹!"# + 𝛼 𝐴!"# − 𝐹!"#

OR
Ft = αAt-1 + (1- α)Ft-1
Where:

𝐹! = 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑣𝑎𝑙𝑢𝑒 𝑓𝑜𝑟 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡

𝐹!"# = 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑣𝑎𝑙𝑢𝑒 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 − 1
𝐴!"# = 𝐴𝑐𝑡𝑢𝑎𝑙 𝑜𝑐𝑐𝑢𝑟𝑒𝑛𝑐𝑒 𝑓𝑜𝑟 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 − 1
𝛼 = 𝐴𝑙𝑝ℎ𝑎, 𝑡ℎ𝑒 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 : Determines how reactive
your forecasts are to actual demand changes
F2 = Y1 (to initiate the computations)

3- 25
 Exponential Smoothing Example
Week Demand Forecast
1 820 820
2 775 820 𝐹$ = 𝐹# + 𝛼 𝐴# − 𝐹# =820 + 0.2(820-820)
3 680 811
4 655 785 𝐹% = 𝐹& + 𝛼 𝐴& − 𝐹& =811+ 0.2(680-811)
5 750 759
6 802 757 𝐹' = 𝐹( + 𝛼 𝐴( − 𝐹( =759+ 0.2(750-759)
7 798 766
8 689 772 𝐹) = 𝐹* + 𝛼 𝐴* − 𝐹* =766+ 0.2(798-766)
9 775 756
10 760 𝐹#+ = 𝐹, + 𝛼 𝐴, − 𝐹, =756+ 0.2(775-756)

3- 26
Exponential Smoothing Problem (1) Data

Week Demand
1 820 Question: Given the weekly
2 775 demand data, what are the
3 680 exponential smoothing
4 655 forecasts for periods 2-10
5 750 using a=0.2?
6 802 Assume F1=D1
7 798
8 689
9 775
10
3- 27
 Exponential Smoothing Problem (1)
Week Demand Forecast
1 820 820
2 775 820 𝐹$ = 𝐹# + 𝛼 𝐴# − 𝐹# =820 + 0.2(820-820)
3 680 811
4 655 785 𝐹% = 𝐹& + 𝛼 𝐴& − 𝐹& =811+ 0.2(680-811)
5 750 759
6 802 757 𝐹' = 𝐹( + 𝛼 𝐴( − 𝐹( =759+ 0.2(750-759)
7 798 766
8 689 772 𝐹) = 𝐹* + 𝛼 𝐴* − 𝐹* =766+ 0.2(798-766)
9 775 756
10 760 𝐹#+ = 𝐹, + 𝛼 𝐴, − 𝐹, =756+ 0.2(775-756)

3- 28
The following table shows respective forecasts for alpha=0.1 and 0.6,
respectively. Note that you can only forecast one time period into the future.

Week Demand 0.1 0.6

1 820 820.00 820.00
2 775 820.00 820.00
3 680 815.50 793.00
4 655 801.95 725.20
5 750 787.26 683.08
6 802 783.53 723.23
7 798 785.38 770.49
8 689 786.64 787.00
9 775 776.88 728.20
10 776.69 756.28 3- 29
Exponential Smoothing Problem (1) Plotting

Note how that the smaller alpha results in a smoother line in this example .
Forecasts are less reactive to demand changes in that case.

900
800 Demand
Demand

700 0.1
600 0.6
500
1 2 3 4 5 6 7 8 9 10
Week

3- 30
 Which alpha to choose?
q UseMAD or Se, to decide which alpha has less
error.
q MAD= Mean Absolute Deviation.
q Se = Standard Error.
q See Performance Measures.xls

3- 31
The MAD Statistic to Determine Forecasting Error

n
1 MAD » 0.8 standard deviation
åA
t=1
t - Ft
1 standard deviation » 1.25 MAD
MAD =
n

q The ideal MAD is zero which would mean

there is no forecasting error

q The larger the MAD, the less the accurate

the resulting model

3- 32
Classification of Time Series Models
q No Trend; No Seasonality
¤ Moving Average Techniques
n Simple MA
n Weighted Moving Average
¤ Exponential Smoothing Technique
n MA, ES,ES with Trend
¤ MAD and Se
q With Trend but Not Seasonality
¤ Exponential Smoothing With Trend
¤ Simple Linear Regression (least Squares method)
q With Trend and Seasonality
¤ Additive vs Multiplicative Demand Models

3- 33
Exponential Smoothing – Effect of Trends
q The presence of a trend in the data causes the
exponential smoothing forecast to always lag
behind the actual data
q This can be corrected by adding a trend adjustment
¤ The trend smoothing constant is delta (𝛿)
𝐹𝐼𝑇! = 𝐹! + 𝑇!
𝐹! = 𝐹𝐼𝑇!"# + 𝛼 𝐴!"# − 𝐹𝐼𝑇!"#
𝑇! = 𝑇!"# + 𝛿 𝐹! − 𝐹𝐼𝑇!"#
𝐹! = 𝐸𝑥𝑝𝑜𝑛𝑒𝑛𝑡𝑖𝑎𝑙𝑙𝑦 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝑇! = 𝐸𝑥𝑝𝑜𝑛𝑒𝑛𝑡𝑖𝑎𝑙𝑙𝑦 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑 𝑡𝑟𝑒𝑛𝑑 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝐹𝐼𝑇! = 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑖𝑛𝑐𝑙𝑢𝑑𝑖𝑛𝑔 𝑡𝑟𝑒𝑛𝑑 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝐹𝐼𝑇!"# = 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑖𝑛𝑐𝑙𝑢𝑑𝑖𝑛𝑔 𝑡𝑟𝑒𝑛𝑑 𝑓𝑜𝑟 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑝𝑒𝑟𝑖𝑜𝑑
𝐴!"# = 𝐴𝑐𝑡𝑢𝑎𝑙 𝑜𝑐𝑐𝑢𝑟𝑒𝑛𝑐𝑒 𝑓𝑜𝑟 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑝𝑒𝑟𝑖𝑜𝑑
𝛼 = 𝑆𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (𝑎𝑙𝑝ℎ𝑎)
𝛿 = 𝑆𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (𝑑𝑒𝑙𝑡𝑎) 3- 34
Example – Exponential Smoothing with
Trend Adjustment
q Calculate the new forecast, assuming the following
¤ The previous forecast including trend (𝐹𝐼𝑇DEF) is 110 and
the previous estimate of the trend (𝑇DEF) is 10
¤ 𝛼 = 0.2 and 𝛿 = 0.3
¤ Actual demand for period t-1 was 115

𝐹D = 𝐹𝐼𝑇DEF + 𝛼 𝐴DEF − 𝐹𝐼𝑇DEF =110+0.2(115-110)=111.0

𝑇D = 𝑇DEF + 𝛿 𝐹D − 𝐹𝐼𝑇DEF =10+0.3(111-110)=10.3
𝐹𝐼𝑇D = 𝐹D + 𝑇D =111.0+10.3=121.3

Objective Question 7, from the back of book.

3- 35
Choosing Alpha and Delta
q Relatively small values for 𝛼 and 𝛿 are common
q Usually in the range 0.1 to 0.3
q𝛼 depends upon how much random variation is
present
q 𝛿 depends upon how steady the trend is
q Measurements of forecast error can be used to
select values of 𝛼 and 𝛿 to minimize overall
forecast error

3- 36
Linear Regression Analysis
q Regression is used to identify the functional
relationship between two or more correlated
variables, usually from observed data
q One variable (the dependent variable) is predicted
for given values of the other variable (the
independent variable)
q Linear regression is a special case which assumes the
relationship between the variables can be explained
with a straight line
𝑌! − 𝑡ℎ𝑒 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒 𝑣𝑎𝑙𝑢𝑒, (Sales)
Yt=a + bt 𝑎 − 𝑡ℎ𝑒 𝑦 − 𝑖𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑙𝑖𝑛𝑒
𝑏 − 𝑡ℎ𝑒 𝑠𝑙𝑜𝑝𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑙𝑖𝑛𝑒
𝑡 − 𝑖𝑛𝑑𝑒𝑥 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑
3- 37
Example 3.2 – Least Squares Method
Quarter Sales Quarter Sales
The least squares method determines the
1 600 7 2,600
parameters a and b such that the sum of
the squared errors is minimized – the 2 1,550 8 2,900
“least squares” 3 1,500 9 3,800
4 1,500 10 4,500
5 2,400 11 4,000
6 3,100 12 4,900

𝑆𝑢𝑚 𝑜𝑓 𝑠𝑞𝑢𝑎𝑟𝑒𝑑 𝑒𝑟𝑟𝑜𝑟𝑠

(𝑦# − 𝑌# )$ +(𝑦$ − 𝑌$ )$ + ⋯ + (𝑦#$ − 𝑌#$ )$

3- 38
 Example 3.2 - Calculations
∑ 𝑡𝑦 − 𝑛𝑡̅ 6 𝑦# 268,200 − 12 ∗ 6.5 ∗ 2,779.2 𝒕 𝒚 𝒕×𝒚 𝒕𝟐 𝒀
𝑏= = = 359.6
∑ 𝑡 ! − 𝑛𝑡̅ ! 650 − 12 ∗ 6.5! 1 600 600 1 801.3
a = 𝑦# − 𝑏𝑡̅ = 2,779.2 − 359.6 ∗ 6.5 = 441.67 2 1,550 3,100 4 1,160.9

3 1,500 4,500 9 1,520.5

𝑌" = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 1 = 801.3
4 1,500 6,000 16 1,880.1

5 2,400 12,000 25 2,239.7

𝑌! = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 2 = 1,160.9
6 3,100 18,600 36 2,599.4
⋮
7 2,600 18,200 49 2,959.0
𝑌"! = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 12 = 4,757.1
8 2,900 23,200 64 3,318.6
The forecast is then extended to periods 13-16
9 3,800 34,200 81 3,678.2
𝑌"# = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 13 = 5,116.4 10 4,500 45,000 100 4,037.8

𝑌"$ = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 14 = 5,476.0 11 4,000 44,000 121 4,397.4

12 4,900 58,800 144 4,757.1

𝑌"% = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 15 = 5,835.6
Sum 78 33,350 268,200 650
𝑌"& = 𝑎 + 𝑏𝑡 = 441.67 + 359.6 ∗ 16 = 6,195.2
Average 6.5 2779.2
t-bar y-bar
3- 39
Regression with Excel
q Microsoft
Excel includes
data analysis
tools, which
can perform
least squares
regression on
a data set

3- 40
Classification of Time Series Models
q No Trend; No Seasonality
¤ Moving Average Techniques
n Simple MA
n Weighted Moving Average
¤ Exponential Smoothing Technique
q With Trend but Not Seasonality
¤ Exponential Smoothing With Trend
¤ Simple Linear Regression (least Squares method)
q With Trend and Seasonality
¤ Additive vs Multiplicative Demand Models

3- 41
Time Series Decomposition
¨ Chronologically ordered data is referred to as a
time series
¨ A time series may contain one or many elements
¤ Trend, seasonal, cyclical, autocorrelation, and random
¨ Identifying these elements and separating the time
series data into these components is known as
decomposition

3- 42
Seasonal Variation
¨ Seasonal variation may be either additive or
multiplicative (shown here with a changing trend)

3- 43
Forecasting with Trend and Seasonal Components
q Additive Seasonal Time Series Model
Forecast including trend and Seasonality = Yt=Tt +St
q Multiplicative Seasonal Time Series Model
Forecast including trend and Seasonality = Yt=Tt X St
See above figures.
Mostly we experience multiplicative models.
q If we are given the actual demand data and a Trend line as in
exhibit 3.10, see example 3.4 and Problem 28 Solutions
provided in the Solutionsto Suggested Problems from Forecasting
Chapter.pdf to see how to find the seasonality indices and
making forecasts.
q Else, if we are given the actual demand data and are asked to
decompose the data into its multiplicative trend and seasonality
components and then make forecasts, follow the next
procedure.

3- 44
Decomposition Using Least Squares
Regression
1. Decompose the time series into its components
a. Find seasonal component
b. Deseasonalize the demand
c. Find the trend component
2. Forecast future values of each component
a. Project trend component into the future
b. Multiply the trend component by the seasonal
component

3- 45
 Multiplicative Time Series Model
Decomposition Steps of Multiplicative Time Series Model
step 1 = avg. demand for the entire data

step 2 = compute avg. demand per season

step 3 = divide avg. seasonal dem. by avg. dem. to get seasonal indices

step 4 = de-seasonalize the data: divide the actuals by its seasonal index

step 5 = apply regression to the de-seasonalized data

step 6 = re-seasonalize the regr. Predictions: Y(regr)*Seasonal Index

See example Problem 28 in Solutions to Suggested

Problems from Forecasting Chapter.pdf.
3- 46
 (Column 1)2
Decomposition -Steps 1 toColumn
4 3÷ Column 1 x
Exhibit 3.11 Column 5 Column 6
(1) (2) (3) (4) (5) (6) (7) (8)
Period Quarter Actual Average of Same Quarters of Seasonal Deseasonalized 𝒕𝟐 𝒕×𝒚𝒅
(t) Demand (y) Each Year (Step 2) Factor Demand (𝒚𝒅) Step 4)
1 I 600 600 + 2,400 + 3,800 0.82 600 1
= 2,266.7 = 735.7 735.70
3 .82
2 II 1,550 1,550 + 3,100 + 4,500 1.10 1,550 4
= 3,050 = 1,412.4 2,824.8
3 1.10
3 III 1,500 1,500 + 2,600 + 4,000 0.97 1,500 9
= 2,700 = 1,544.0 4,632
3 0.97
4 IV 1,500 1,500 + 2,900 + 4,900 1.12 1,500 16
= 3,100 = 1,344.8 5,379
3 1.12

5 I 2,400 Step1) 0.82 2,942.6 25 14,713

6 II 3,100 (2266.7+3050+2700+3100)/4= 1.10 2,824.7 36 16,948
2779.18
7 III 2,600 0.97 2,676.2 49 18,733
8 IV 2,900 Step 3) 1.12 2,599.9 64 20,799
2266.7/2779.18= 0.8156 = 0.82
9 I 3,800 3050/2779.18 =1.0975 = 1.10 0.82 4,659.2 81 41,933
10 II 4,500 2700/2779.18 =0.9715 = 0.97 1.10 4,100.4 100 41,004
3100/2779.18 =1.1154= 1.12
11 III 4,000 0.97 4,117.3 121 45,290
12 IV 4,900 1.12 4,392.9 144 52,715

Step1:Average Demand = (600+1550+1500+…+4900)/12=2779.17 (rounding error above) 3- 47

 Decomposition – Steps 5
q Develop a least squares regression line for the deseasonalized
data: Either manually or using Excel.
q Project the regression line through the period of the forecast

Regression Results:
Y = 555.0 + 342.2t

Forecast for
periods 13-16

3- 48
 Decomposition – Step 6
q Create the final forecast by adjusting the regression line by the
seasonal factor
Y = 555.0 + 342.2t
Y = 555 + 342.2(13)= 5003.5 x 0.82 = 4102.87
Y = 555 + 342.2(14)= 5345.8 x 1.10 = 5880.27
Y = 555 + 342.2(15)= 5687.9 x 0.97 =5517.26
Y = 555 + 342.2(16)= 6030.1 x 1.12 = 6753.71

Period Quarter Y from Regression Seasonal Factor Forecast (F x

Seasonal Factor
13 I 5,003.5 0.82 4,102.87
14 II 5,345.7 1.10 5,880.27
15 III 5,687.9 0.97 5,517.26
16 IV 6,030.1 1.12 6,753.71

3- 49
Forecast Errors
q Forecast error is the difference between the forecast
value and what actually occurred
q All forecasts contain some level of error
q Sources of error
¤ Bias – when a consistent mistake is made
¤ Random – errors that are not explained by the model being
used
q Measures of error
¤ Mean absolute deviation (MAD)
¤ Mean absolute percent error (MAPE)
¤ Tracking signal

3- 50
 Forecast Error Measurements
q Ideally, MAD will be zero q MAPE scales the forecast error to
(no forecasting error) the magnitude of demand
q Larger values of MAD
MAD
indicate a less accurate MAPE =
Average Demand
model
∑/-.# 𝐴! − 𝐹! q Tracking signal indicates whether
MAD =
𝑛 forecast errors are accumulating
𝑤ℎ𝑒𝑟𝑒
𝑡 = period number over time or not (either positive or
𝐴! = actual demand during period 𝑡 negative errors)
𝐹! = forecast demand during period 𝑡
Running sum of forecast errors
𝑛 = total number of periods 𝑇𝑆 =
Mean absolute deviation

3- 51
 Computing Forecast Error

Month Forecast Actual Deviation RSFE Abs. Sum of MAD TS = RSFE

(error) Dev. Abs. Dev. (MAE) /MAD
1 1,000 950 -50 -50 50 50 50 -1
2 1,000 1,070 +70 +20 70 120 60 0.33
3 1,000 1,100 +100 +120 100 220 73.3 1.64
4 1,000 960 -40 +80 40 260 65 1.2
5 1,000 1,090 +90 +170 90 350 70 2.4
6 1,000 1,050 +50 +220 50 400 66.7 3.3
Overall
400 66.7 220
MAD = = 66.7 MAPE = = 6.43% TS = = 3.3
6 1036.7 66.7

3- 52
Causal Relationship Forecasting
q Causalrelationship forecasting uses independent
variables other than time to predict future demand
¤ This independent variable must be a leading indicator
q Many apparently causal relationships are actually
just correlated events – care must be taken when
selecting causal variables

3- 53
Multiple Regression Techniques

q Often, more than one independent variable may be

a valid predictor of future demand
q In this case, the forecast analyst may utilize multiple
regression
¤ Analogous to linear regression analysis, but with
multiple independent variables
¤ Multiple regression is supported by statistical software
packages

3- 54
Multiple Regression analysis
q Identify factors (independent variables) that can be used to
predict the values for the forecast variable (e.g., sales).
q A little more involved data collection than the time series
cases.
q Use Excel (Tools/Data analysis) to obtain the statistics.
q Check each independent variable and the intercept for
statistical significance (p-values ~ £ 0.05)
q Drop insignificant variable(s), one at a time and re-run the
model as many times as needed
q If the “clean” model has a good adjusted R2 (subjective
measure) the final model can be used to make decisions.

q See the solution for Problem 29 in the Solutions to Suggested

Problems from Forecasting Chapter.pdf.
3- 55
Qualitative Forecasting Techniques
q Generally used to take advantage of expert
knowledge
q Useful when judgment is required, when products
are new, or if the firm has little experience in a new
market
q Examples
¤ Market research
¤ Panel consensus
¤ Historical analogy
¤ Delphi method

3- 56
Collaborative Planning, Forecasting,
and Replenishment (CPFR)
qA web-based process used to coordinate the efforts
of a supply chain
¤ Demand forecasting
¤ Production and purchasing

¤ Inventory replenishment

q Integratesall members of a supply chain –

manufacturers, distributors, and retailers
q Depends upon the exchange of internal information
to provide a more reliable view of demand

3- 57
Things to do before next time.
q Read Chapter 3
q Understand the Ch 3 Solved Problems
q Solve Ch 3 Problems: 6, 7, 15, 17, 18, 22, 27, 28,
and 29
q Read Chapter 8 –Sales and Operations Planning

58